Developer And Task Recommendation In Software Crowdsourcing

Software crowdsourcing has been a new software development model with the prosperity of the Internet,which can utilize the wisdom of the crowd to develop software.It can improve the development efficiency,help the crowdsourcer to save spending,and satisfy the developers to work at home and create more money.But at the same time,there exist problems in software crowdsourcing,such as intelligent matching between developers and tasks,quality control,waste of human resources,and developer incentive problems.These problems have seriously restricted the development of the software crowdsourcing.This paper researches on developer and task recommendation in software crowdsourcing in depth.Exploring the important factors in the recommendations according to the characteristics of the software crowdsourcing,a task model and a developer model in software crowdsourcing are established from multi-dimensions.Then a comprehensive quality based developer recommendation method,SIMSOFT,and a multi-level progressive complementary task recommendation method,SIMFIT,in software crowdsourcing are proposed.They can help to complete tasks better.The contributions of this paper include:1)Developer model and task model in software crowdsourcing are built.The key factors which can affect the recommendation in software crowdsourcing are found by analyzing the characteristics of software crowdsourcing and researching the data of the crowdsourcing platform.Based on these factors,a task model in software crowdsourcing is established from four dimensions,which are basic information,skill,resource and description.And a developer model in software crowdsourcing is established from five dimensions,which are basic information,service scope,skill,guarantee and recent performance.The credit is the core attribute of guarantee,and it can be calculated from their historical information according to reward-related principle,time-related principle,and punishment warning principle.2)A comprehensive quality based developer recommendation method in software crowdsourcing,SIMSOFT,is proposed.Based on historical crowdsourcing data,SIMSOFT takes three factors into account and quantifies them to recommend appropriate developers for a specific task in software crowdsourcing.These three factors are the competency of the developer to the basic requirements of a task,the development history of the developer and the soft power of the developer,separately.When calculating the similarity between a certain task and the historical tasks of a developer,Jaccard Coefficient is improved and it takes software category,sdlc,skill and budget as input;when calculating the soft power of a developer,the guarantee and the recent performance of the developer are considered.To verify the effectiveness of SIMSOFT,the data from China's largest crowdsourcing platform,Zhuabajie,is used to finish several experiments.The data analysis and the greedy algorithm are used to tune parameters.Then the compared experiments are carried on.Experimental results show that SIMSOFT can achieve effective developer recommendation: 10%-precision can reach 80.92%,20%-precision can reach 94.74%.It is significantly better than the existing method CBR.3)A multi-level progressive complementary task recommendation method in software crowdsourcing,SIMFIT,is proposed.It utilizes the match between the developer and the task,the similarity between the task and the historical tasks of the developer,and the new participation of the similar developers,to recommend appropriate tasks to a specific developer.To verify the effectiveness of SIMFIT,using the data from Zhubajie,the greedy algorithm is used on training set to finish parameter tuning.And then the compared experiments are carried on.The experimental results show that SIMFIT can achieve effective task recommendation: the average precision rate,the average recall rate and the average F-measure can reach 0.4403,0.9833 and 0.6082 separately.It outperforms the general crowdsourcing task recommendation method CBAM.